Persistent memory for AI coding agents: facts with provenance, superseding stale knowledge, and MCP server.
by dat999zx
Knowl is a persistent memory layer for AI coding agents. It stores typed knowledge atoms — decisions, constraints, architecture, facts, goals, state, and skills — with provenance (observed, stated, inferred). When a fact is superseded (e.g., "rate limit changed from 100 to 500"), the predecessor is retired, not deleted, keeping a full timeline.
The tool runs as an MCP server, so any MCP-compatible client works. Installation is npm install -g @dat999zx/knowl. Knowledge stays local to your repository until you publish; embeddings and session history are never uploaded.
The live site explains the problem of stale knowledge and positions Knowl as a solution that outlives team members. It includes performance benchmarks and a list of supported clients. This is infrastructure for teams that want their agents to remember past decisions.
Built with
Model + tools used
Last checked 2026-09-01 11:44 UTC
Open live app ↗| timestamp | result | latency |
|---|---|---|
| 2026-09-01 11:44 UTC | live | 45ms |
| 2026-08-26 21:22 UTC | live | 53ms |
| 2026-08-26 20:16 UTC | live | 121ms |